Bidirectional Sequence Classification for Named Entities Recognition

Abstract

With this paper is presented a system for Named Entities Recognition, based on the Perceptron Algorithm. In the proposed frame- work, the order of the inference is not forced into a monotonic behavior (left-to-right), but is learned together with the parameters of the local classifier. The system tested on the task of Italian NER at EVALITA 2009 obtained the second position, with an F1 measure of 81.46%.